< Docs
  • Python


This operation inserts new records into the database or updates existing ones.


An upsert is a data-level operation that will overwrite an existing entity if a specified field already exists in a collection, and insert a new entity if the specified value doesn’t already exist.

Request Syntax

    data: List | pandas.DataFrame | Dict,, 
    timeout=float | None
from pymilvus import Collection, Partition

# Get an existing collection
collection = Collection(name="string")

# Get an existing partition
partition = Partition(name="string")

# Prepare your data
data = ...

# Upsert data into partition


  • data (list | dict | pandas.DataFrame) -


    The data to insert into the current collection.

    The data to insert should match the schema of the current collection. You can organize your data into:

    • A list of columns

      Each column is a list of values of all entities in that column.

      data = [
          [0,1,2,3,4],                         # id
          [                                    # vector
    • A pandas.DataFrame

      You can form a data frame in any way, as demonstrated in the Example section on this page.

      data = pd.DataFrame({
          "id": [5,6,7,8,9],
          "vector": [
    • A list of rows or just a row

      Each row is a dictionary that represents an entity.

      data = [
          {"id": 10, "vector": [0.1,0.2,-0.3,-0.4,0.5]},
          {"id": 11, "vector": [0.3,-0.1,-0.2,-0.6,0.7]},
          {"id": 12, "vector": [-0.6,-0.3,0.2,0.8,0.7]},
          {"id": 13, "vector": [0.6,0.2,-0.3,-0.8,0.5]},
          {"id": 14, "vector": [0.3,0.1,-0.2,-0.6,-0.7]},
      # or 
      data = {"id": 15, "vector": [0.3,0.1,-0.2,-0.6,-0.7]},
  • timeout (float | None)

    The timeout duration for this operation. Setting this to None indicates that this operation timeouts when any response arrives or any error occurs.




A MutationResult object that contains the following fields:

  • insert_count (int)

    The count of inserted entities.

  • delete_count (int)

    The count of deleted entities.

  • upsert_count (int)

    The count of upserted entities.

  • succ_count (int)

    The count of successful executions during this operation.

  • succ_index (list)

    A list of index numbers starting from 0, each indicating a successful operation.

  • err_count (int)

    The count of failed executions during this operation.

  • err_index (list)

    A list of index numbers starting from 0, each indicating a failed operation.

  • primary_keys (list)

    A list of primary keys for the inserted entities.

  • timestamp (int)

    The timestamp at which this operation is completed.


  • MilvusException

    This exception will be raised when any error occurs during this operation.


from pymilvus import Collection, Partition, FieldSchema, CollectionSchema, DataType

# Define collection schema    
schema = CollectionSchema([
    FieldSchema("film_id", DataType.INT64, is_primary=True),
    FieldSchema("films", dtype=DataType.FLOAT_VECTOR, dim=2)

# Get an existing collection
collection = Collection("test_partition_insert", schema)

# Get an existing partition in the current collection
partition = Partition(collection, "comedy", "comedy films")

# Prepare the data to insert
data = [
    [i for i in range(10)],
    [[float(i) for i in range(2)] for _ in range(10)]

# Upsert data
res = partition.upsert(data)

# Return the count of upserted entities